- vector space/Linear space 中,向量之间的运算只有加法和数(统称线性运算),但是,如果以解析几何中的三维空间 R 3 \mathbb{R}^3 R3 作为线性空间的一个模型,我们会发现 R 3 \mathbb{R}^3 R3 诸如向量长度、向量夹角等度量概念无法在 Linear space 中得到反映。为此,有必要在一般的 Linear space 中引进 inner product 內积运算,这样就导出了 Inner product spaces 內积空间的概念
1. Inner product spaces
1.1 inner product (內积)
-
Definition:Let V V V a vector space over filed P P P, α , β ∈ V \alpha,\beta\in V α,β∈V. If there is an operation ( ⋅ , ⋅ ) (·,·) (⋅,⋅) from V V V to P P P satisfying the following rules
- 对称性: ( α , β ) = ( β , α ) (\alpha,\beta) = (\beta,\alpha) (α,β)=(β,α)
- 可加性: ( α + β , γ ) = ( α , γ ) + ( β , γ ) (\alpha+\beta,\gamma) = (\alpha,\gamma)+(\beta,\gamma) (α+β,γ)=(α,γ)+(β,γ)
- 齐次性: ( k α , β ) = k ( α , β ) (k\alpha,\beta) = k(\alpha,\beta) (kα,β)=k(α,β)
- 非负性: ( α , α ) ≥ 0 (\alpha,\alpha)\geq 0 (α,α)≥0, α = 0 ⇔ ( α , α ) = 0 \alpha=0 \Leftrightarrow (\alpha,\alpha)=0 α=0⇔(α,α)=0
then ( α , β ) (\alpha,\beta) (α,β) is the
inner product
of α \alpha α and β \beta β -
用 V V V 的一组基来表示 α , β \alpha,\beta α,β, 利用上述性质把连加的内积拆开,有
其中 y H y^H yH 是向量 y y y 的共轭转置。可见矩阵 A A A 和使用的基 ε i \varepsilon_i εi 是一一对应的,当基选用标准正交基时, A A A 矩阵最简单,为单位阵,详见第 2 节
1.2 inner product space (內积空间)
- 內积空间其实就是向量空间/线性空间的基础上加了一个內积运算
- Definition:Let
V
V
V a vector space over filed
P
P
P with inner product operation, then
V
V
V is called an
inner product space
on P P P - Special case
- if
P
=
R
P=R
P=R,
V
V
V is called
Euclid space
(欧氏空间) - if
P
=
C
P=C
P=C,
V
V
V is called
Unitary space
(酉空间)
- if
P
=
R
P=R
P=R,
V
V
V is called
- In inner product space, the following properties hold
- ( α , β + γ ) = ( α , β ) + ( α , γ ) (\alpha,\beta+\gamma) = (\alpha,\beta)+(\alpha,\gamma) (α,β+γ)=(α,β)+(α,γ)
- ( α , k β ) = k ˉ ( α , β ) (\alpha,k\beta) = \bar{k}(\alpha,\beta) (α,kβ)=kˉ(α,β)(从第二个元提常数需要变成共轭, k ˉ \bar{k} kˉ 是 k k k 的共轭)
- ( α , 0 ) = ( 0 , α ) = 0 (\alpha,\pmb{0}) = (\pmb{0},\alpha) = 0 (α,000)=(000,α)=0
1.3 More Notion in inner product space
1.3.1 Length of vector
-
Definition:Suppose that V V V is an inner product space, The
length
of vector α ∈ V \alpha \in V α∈V is
∥ α ∥ = ( α , α ) \Vert \alpha \Vert = \sqrt{(\alpha,\alpha)} ∥α∥=(α,α)
The vector of length 1 is called aunit
vector -
For any α , β ∈ V \alpha,\beta\in V α,β∈V, the length ∥ α ∥ \Vert \alpha \Vert ∥α∥ satisfies the following properties
- ∥ α ∥ ≥ 0 \Vert\alpha\Vert\geq 0 ∥α∥≥0 and α = 0 ⇔ ∥ α ∥ = 0 \alpha=0 \Leftrightarrow \Vert\alpha\Vert=0 α=0⇔∥α∥=0
- ∥ k α ∥ = ∣ k ∣ ∥ α ∥ \Vert k\alpha\Vert = |k|\Vert\alpha\Vert ∥kα∥=∣k∣∥α∥ for any k ∈ P k\in P k∈P
- ∥ α + β ∥ 2 + ∥ α − β ∥ 2 = 2 ( ∥ α ∥ 2 + ∥ β ∥ 2 ) \Vert \alpha+\beta\Vert^2+\Vert \alpha-\beta\Vert^2 = 2(\Vert\alpha\Vert^2+\Vert\beta\Vert^2) ∥α+β∥2+∥α−β∥2=2(∥α∥2+∥β∥2) for all α , β ∈ V \alpha,\beta\in V α,β∈V
-
∣
(
α
,
β
)
∣
≤
∥
α
∥
∥
β
∥
|(\alpha,\beta)|\leq \Vert\alpha\Vert \Vert\beta\Vert
∣(α,β)∣≤∥α∥∥β∥(
Cauchy Inequality 柯西不等式
) -
∥
α
+
β
∥
≤
∥
α
∥
+
∥
β
∥
∥
α
−
β
∥
≤
∥
α
∥
+
∥
β
∥
\begin{aligned}\Vert\alpha+\beta\Vert \leq \Vert\alpha\Vert+\Vert\beta\Vert\\\Vert\alpha-\beta\Vert \leq \Vert\alpha\Vert+\Vert\beta\Vert\end{aligned}
∥α+β∥≤∥α∥+∥β∥∥α−β∥≤∥α∥+∥β∥ (
三角不等式
)
-
上面第 4 条 Cauchy Inequality 是一个重要的不等式
- 证明,构造一个向量
α
+
x
β
\alpha+x\beta
α+xβ,其中
x
=
−
(
α
,
β
)
(
β
,
β
)
x=-\frac{(\alpha,\beta)}{(\beta,\beta)}
x=−(β,β)(α,β),利用自己和自己做内积的非负性,有
∵ ( α + k β , α + k β ) = ( α , α ) + x 2 ( β , β ) + 2 x ( α , β ) = ( α , α ) + ( α , β ) 2 ( β , β ) − 2 ( α , β ) 2 ( β , β ) ≥ 0 ∴ ( α , α ) ( β , β ) ≥ ( α , β ) 2 ∴ ∥ α ∥ 2 ∥ β ∥ 2 ≥ ( α , β ) 2 ∴ ∥ α ∥ ∥ β ∥ ≥ ∣ ( α , β ) ∣ \begin{aligned} &\begin{aligned} \because(\alpha+k\beta,\alpha+k\beta) &= (\alpha,\alpha)+x^2(\beta,\beta)+2x(\alpha,\beta) \\ &=(\alpha,\alpha)+\frac{(\alpha,\beta)^2}{(\beta,\beta)} - 2\frac{(\alpha,\beta)^2}{(\beta,\beta)} \geq 0 \end{aligned} \\ &\therefore (\alpha,\alpha)(\beta,\beta) \geq (\alpha,\beta)^2 \\ &\therefore \Vert\alpha\Vert^2\Vert\beta\Vert^2 \geq (\alpha,\beta)^2 \\ &\therefore \Vert\alpha\Vert\Vert\beta\Vert \geq \big|(\alpha,\beta)\big| \\ \end{aligned} ∵(α+kβ,α+kβ)=(α,α)+x2(β,β)+2x(α,β)=(α,α)+(β,β)(α,β)2−2(β,β)(α,β)2≥0∴(α,α)(β,β)≥(α,β)2∴∥α∥2∥β∥2≥(α,β)2∴∥α∥∥β∥≥∣∣(α,β)∣∣ - 在不同的内积空间中,柯西不等式有不同的表现形式
- 证明,构造一个向量
α
+
x
β
\alpha+x\beta
α+xβ,其中
x
=
−
(
α
,
β
)
(
β
,
β
)
x=-\frac{(\alpha,\beta)}{(\beta,\beta)}
x=−(β,β)(α,β),利用自己和自己做内积的非负性,有
-
uniformization归一化 / unitization单位化
:For any α ≠ 0 \alpha\neq 0 α=0, α ∥ α ∥ \frac{\alpha}{\Vert\alpha\Vert} ∥α∥α is a unit vector
1.3.2 Distance of vectors
- Definition:if
V
V
V is an inner product space, the
distance
of α , β ∈ V \alpha,\beta\in V α,β∈V is defined as
d ( α , β ) = ∥ α − β ∥ d(\alpha,\beta) = \Vert \alpha-\beta\Vert d(α,β)=∥α−β∥ - the distance
d
(
α
,
β
)
d(\alpha,\beta)
d(α,β) satisfies the following rules
- 非负性: d ( α , β ) ≥ 0 d(\alpha,\beta)\geq 0 d(α,β)≥0 and d ( α , β ) = 0 ⇔ α = β d(\alpha,\beta)= 0 \Leftrightarrow \alpha=\beta d(α,β)=0⇔α=β
- 对称性: d ( α , β ) = d ( β , α ) d(\alpha,\beta) = d(\beta,\alpha) d(α,β)=d(β,α)
- 三角不等式: d ( α , γ ) ≤ d ( α , β ) + d ( β , γ ) d(\alpha,\gamma)\leq d(\alpha,\beta)+d(\beta,\gamma) d(α,γ)≤d(α,β)+d(β,γ)
1.3.3 Angle of vectors
- Definition:Suppose that
V
V
V is an inner product space,
α
,
β
∈
V
,
α
,
β
≠
0
\alpha,\beta\in V, \alpha,\beta \neq \pmb{0}
α,β∈V,α,β=000, the
aggle
of α \alpha α and β \beta β is defined as
< α , β > = a r g c o s ( α , β ) ∥ α ∥ ∥ β ∥ , 0 ≤ < α , β > ≤ π \big<\alpha,\beta\big> = \mathrm{argcos} \frac{(\alpha,\beta)}{\Vert\alpha\Vert\Vert\beta\Vert}, 0 \leq \big<\alpha,\beta\big> \leq \pi ⟨α,β⟩=argcos∥α∥∥β∥(α,β),0≤⟨α,β⟩≤π - Especially if
(
α
,
β
)
=
0
(\alpha,\beta)=0
(α,β)=0, we call that
α
\alpha
α is
orthogonal
to β \beta β, denote as ( α , β ) = 0 ⇔ α ⊥ β (\alpha,\beta)=0 \Leftrightarrow \alpha\perp\beta (α,β)=0⇔α⊥β
Two orthogonal vectors satisfy thePythagorean Law
(勾股定理)
α ⊥ β ⇒ ∥ α + β ∥ 2 = ∥ α − β ∥ 2 = ∥ α ∥ 2 + ∥ β ∥ 2 \alpha \perp \beta \Rightarrow \Vert\alpha+\beta\Vert^2 =\Vert\alpha-\beta\Vert^2 = \Vert\alpha\Vert^2+\Vert\beta\Vert^2 α⊥β⇒∥α+β∥2=∥α−β∥2=∥α∥2+∥β∥2
1.4 Example
1.4.1 Example 1
- There are vectors
x
=
(
x
1
,
x
2
,
.
.
.
,
x
n
)
T
,
y
=
(
y
1
,
y
2
,
.
.
.
,
y
n
)
T
x=(x_1,x_2,...,x_n)^T,y=(y_1,y_2,...,y_n)^T
x=(x1,x2,...,xn)T,y=(y1,y2,...,yn)T, define the inner product operation as
( x , y ) = y T x = ∑ i = 1 n x i y i x , y ∈ R n ( x , y ) = y H x = ∑ i = 1 n x i y i ˉ x , y ∈ C n \begin{aligned} &(x,y) = y^Tx = \sum_{i=1}^nx_iy_i \space\space\space x,y \in R^n \\ &(x,y) = y^Hx = \sum_{i=1}^nx_i\bar{y_i} \space\space\space x,y \in C^n \end{aligned} (x,y)=yTx=i=1∑nxiyi x,y∈Rn(x,y)=yHx=i=1∑nxiyiˉ x,y∈Cn
then R n R^n Rn and C n C^n Cn are inner product spaces
1.4.2 Example 2
- Vector space
C
[
a
,
b
]
C[a,b]
C[a,b] is an inner product space with the inner product operation defined as
( f , g ) = ∫ a b f ( x ) g ( x ) d x (f,g) = \int_a^bf(x)g(x)dx (f,g)=∫abf(x)g(x)dx - Proof:
- ( f , g ) = ∫ a b f ( x ) g ( x ) d x = ∫ a b g ( x ) f ( x ) d x = ( g , f ) (f,g) = \int_a^bf(x)g(x)dx = \int_a^bg(x)f(x)dx = (g,f) (f,g)=∫abf(x)g(x)dx=∫abg(x)f(x)dx=(g,f)
- ( k f , g ) = ∫ a b k f ( x ) g ( x ) d x = k ∫ a b f ( x ) g ( x ) d x = k ( f , g ) (kf,g) = \int_a^bkf(x)g(x)dx = k\int_a^bf(x)g(x)dx = k(f,g) (kf,g)=∫abkf(x)g(x)dx=k∫abf(x)g(x)dx=k(f,g)
- ( f + g , h ) = ∫ a b ( f ( x ) + g ( x ) ) h ( x ) d x = ∫ a b f ( x ) h ( x ) d x + ∫ a b g ( x ) h ( x ) d x = ( f , h ) + ( g , h ) (f+g,h) = \int_a^b(f(x)+g(x))h(x)dx = \int_a^bf(x)h(x)dx+\int_a^bg(x)h(x)dx = (f,h)+(g,h) (f+g,h)=∫ab(f(x)+g(x))h(x)dx=∫abf(x)h(x)dx+∫abg(x)h(x)dx=(f,h)+(g,h)
- ( f , f ) = ∫ a b ( f ( x ) ) 2 d x ≥ = 0 (f,f) = \int_a^b(f(x))^2dx \geq=0 (f,f)=∫ab(f(x))2dx≥=0, and ∫ a b ( f ( x ) ) 2 d x = 0 ⇔ f ( x ) ≡ 0 \int_a^b(f(x))^2dx =0 \Leftrightarrow f(x) \equiv 0 ∫ab(f(x))2dx=0⇔f(x)≡0
1.4.3 Example 3
- Consider the vector space
C
[
a
,
b
]
C[a,b]
C[a,b], the convolution operation
( f , g ) = ∫ a b f ( x ) g ( a + b − x ) d x (f,g) = \int_a^bf(x)g(a+b-x)dx (f,g)=∫abf(x)g(a+b−x)dx
is a inner product operation or not? Why? - Answer:No
2. Orthonormal basis
2.1 Orthonormal basis(标准正交基)
2.1.1 Definition
- Suppose that
α
1
,
α
2
,
.
.
.
,
α
m
≠
0
\alpha_1,\alpha_2,...,\alpha_m \neq \pmb{0}
α1,α2,...,αm=000 are nonzero vectors in inner product space
V
V
V. If
(
α
i
,
α
j
)
=
0
,
i
≠
j
(\alpha_i,\alpha_j)=0, i\neq j
(αi,αj)=0,i=j then
α
1
,
α
2
,
.
.
.
,
α
m
\alpha_1,\alpha_2,...,\alpha_m
α1,α2,...,αm are
orthogonal (正交的)
. Especially, if
( α i , α j ) = δ i j = { 0 i ≠ j 1 i = j (\alpha_i,\alpha_j) = \delta_{ij} = \begin{cases} 0& i\neq j\\ 1& i= j \end{cases} (αi,αj)=δij={01i=ji=j
the vector group α 1 , α 2 , . . . , α m \alpha_1,\alpha_2,...,\alpha_m α1,α2,...,αm are calledorthonormal (标准正交的)
- In
n
n
n-dimensional inner product space
V
V
V,
n
n
n orthogonal vectors form a basis of
V
V
V which is named
Orthogonal Basis (正交基)
. If the basis is orthogonal and unit, it is calledOrthonormal Basis (标准正交基)
- Suppose that
V
1
,
V
2
V_1, V_2
V1,V2 are two subspaces of inner product space
V
V
V
- Given
α
∈
V
\alpha\in V
α∈V, if
∀
β
∈
V
1
\space \forall \beta \in V_1
∀β∈V1 satisfies
(
α
,
β
)
=
0
(\alpha,\beta)=0
(α,β)=0,
α
\alpha
α is said to be orthogonal to
V
1
V_1
V1, denoted by
α
⊥
V
1
\alpha\perp V_1
α⊥V1
向量正交于空间
,指此向量与空间中任意向量都正交 - For each
α
∈
V
1
\alpha \in V_1
α∈V1 and
β
∈
V
2
\beta \in V_2
β∈V2, if
(
α
,
β
)
=
0
(\alpha,\beta) =0
(α,β)=0,
V
1
V_1
V1 is said to be orthogonal to
V
2
V_2
V2, denoted by
V
1
⊥
V
2
V_1\perp V_2
V1⊥V2
一空间正交于另一空间
,指一空间中任意向量与另一空间中任意向量都正交(只要两个空间的某组基正交)
- Given
α
∈
V
\alpha\in V
α∈V, if
∀
β
∈
V
1
\space \forall \beta \in V_1
∀β∈V1 satisfies
(
α
,
β
)
=
0
(\alpha,\beta)=0
(α,β)=0,
α
\alpha
α is said to be orthogonal to
V
1
V_1
V1, denoted by
α
⊥
V
1
\alpha\perp V_1
α⊥V1
- If two subspace
V
1
,
V
2
V_1,V_2
V1,V2 of inner product space
V
V
V are orthogonal (i.e.
V
1
⊥
V
2
V_1\perp V_2
V1⊥V2), then
V
1
+
V
2
V_1+V_2
V1+V2 is named
orthogonal sum (正交和)
, denoted by V 1 ⊕ V 2 V_1\oplus V_2 V1⊕V2.(正交和一定是直和,见下面 Theorem 3) - Suppose that
V
1
V_1
V1 is a subspace of inner product space
V
V
V, the subspace of
V
V
V including all the vectors orthogonal to
V
1
V_1
V1 is called the
orthogonal complement (正交补)
of V 1 V_1 V1, denoted by
V 1 ⊥ = { α ∈ V ∣ α ⊥ V 1 } V_1^\perp = \{\alpha\in V|\alpha\perp V_1\} V1⊥={α∈V∣α⊥V1}
V 1 ⊥ V_1^\perp V1⊥ 是 V V V 的 subspace,要证明这一点,可以先证明其包含 0 \pmb{0} 000 所以非空,再论证其中元素满足加法和数乘的封闭性 - Suppose that
V
1
V_1
V1 is a subspace of an inner product space
V
V
V. If any given
α
∈
V
\alpha\in V
α∈V can be decomposed as
α
=
α
1
+
α
2
\alpha = \alpha_1+\alpha_2
α=α1+α2
where α 1 ∈ V 1 \alpha_1\in V_1 α1∈V1 and α 2 ⊥ V 1 \alpha_2 \perp V_1 α2⊥V1, then α 1 \alpha_1 α1 is called theorthogonal projection (正交投影)
of α \alpha α on V 1 V_1 V1
由于 V = V 1 ⊕ V 1 ⊥ V = V_1\oplus V_1^\perp V=V1⊕V1⊥ (下面定理 4),任何一个向量都可以分解为相互垂直的两部分,2.2 节的标准正交化过程,其实就是在不停地减去正交投影部分,保留与之前的空间正交的部分 - Assume that
V
1
V_1
V1 is a subspace of an inner product space
V
V
V. Given
α
∈
V
\alpha\in V
α∈V, if there exist a vector
a
1
∈
V
1
a_1\in V_1
a1∈V1 such that
d ( α , V 1 ) : = ∥ α − α 1 ∥ = min β ∈ V 1 ∥ α − β ∥ d(\alpha,V_1):= \Vert\alpha-\alpha_1\Vert = \min_{\beta\in V_1}\Vert\alpha-\beta\Vert d(α,V1):=∥α−α1∥=β∈V1min∥α−β∥
then α 1 \alpha_1 α1 is named theOptimal Approximation (最优近似)
of α \alpha α over V 1 V_1 V1
这里定义了向量到空间的距离为 “该向量到空间中所有向量距离的最小值”
如上所述,subspace V 1 V_1 V1 中距离 α \alpha α 最近的向量是 α \alpha α 在 V 1 V_1 V1 中的 orthogonal projection 正交投影
2.1.2 Theorem
- If the vectors
α
1
,
α
2
,
.
.
.
,
α
m
\alpha_1,\alpha_2,...,\alpha_m
α1,α2,...,αm in inner product space are orthogonal , then
α
1
,
α
2
,
.
.
.
,
α
m
\alpha_1,\alpha_2,...,\alpha_m
α1,α2,...,αm are linearly independent
一组正交向量一定线性无关
- Suppose that
V
V
V is an
n
n
n-dimensional inner product space. The orthonormal basis of
V
V
V surely exist, and arbitrary orthonormal vector group in
V
V
V can be expanded to be an orthonormal basis of
V
V
V
n维内积空间中,标准正交基一定存在,且任意一组标准正交向量可以扩展为一组标准正交基
- 随便给一组基,可以用下面的标准正交化过程变为一组标准正交基
- 给出一组标准正交向量,由于正交,它们一定线性无关,把 V V V 中不能被它们线性表出的向量都加入这组向量,这样就得到了 V V V 的一组基,再用标准正交化过程处理,即得一组标准正交基
- Assume that
V
1
V_1
V1 and
V
2
V_2
V2 are twp subspace of inner product space
V
V
V, if
V
1
⊥
V
2
V_1\perp V_2
V1⊥V2 then
V
1
+
V
2
V_1+V_2
V1+V2 is a direct sum
若內积空间的两个子空间垂直,则两个子空间的和为直和
使用等价条件 V 1 ⋂ V 2 = { 0 } V_1\bigcap V_2 = \{\pmb{0 }\} V1⋂V2={000} 证明,用反证法
- Suppose that inner product space
V
V
V is the finite dimensional,
V
1
V_1
V1 is a subspace of
V
V
V. There exists unique subspace
V
1
⊥
V_1^\perp
V1⊥, such that
V
=
V
1
⊕
V
1
⊥
V = V_1\oplus V_1^\perp
V=V1⊕V1⊥
一个內积空间可以拆成某一子空间及其正交补的正交直和
example: Let α = [ 1 0 0 0 ] ∈ R 2 × 2 \alpha = \begin{bmatrix}1 &0 \\0 &0\end{bmatrix} \in R^{2\times 2} α=[1000]∈R2×2, V 1 = s p a n ( α ) V_1=\mathrm{span}(\alpha) V1=span(α), ( A , B ) = t r ( B H A ) (A,B) = tr(B^HA) (A,B)=tr(BHA), please show V 1 ⊥ V_1^\perp V1⊥ 下面从不同的角度给出两种解法
- (Projection Theorem) If
V
1
V_1
V1 is a finite dimensional subspace of an inner product space
V
V
V, then each
α
∈
V
\alpha\in V
α∈V has the unique orthogonal projection on
V
1
V_1
V1
內积空间中的任意向量,都在某个子空间中有唯一的正交投影
(因为內积空间可以分解为任意子空间与其正交补的正交直和,內积中的任意向量也就可以唯一地直和分解到这两个子空间中)
2.2 Orhogonalization Process(标准正交化过程)
- If the vectors
α
1
,
α
2
,
.
.
.
,
α
m
\alpha_1,\alpha_2,...,\alpha_m
α1,α2,...,αm in inner product space
V
V
V are linearly independent, then there is an
orthonormal vector group
ε 1 , ε 2 , . . . , ε m \varepsilon_1,\varepsilon_2,...,\varepsilon_m ε1,ε2,...,εm which are equivalent (互相可以线性表出) to α 1 , α 2 , . . . , α m \alpha_1,\alpha_2,...,\alpha_m α1,α2,...,αm
内积空间中一组线性无关向量一定可以标准正交化
- 正交化的思想:把每个向量中和它不正交的部分减掉
- 标准化的思想:把向量的长度缩短其模倍
- 将线性无关向量组
α
1
,
α
2
,
.
.
.
,
α
m
\alpha_1,\alpha_2,...,\alpha_m
α1,α2,...,αm 标准正交化为
ε
1
,
ε
2
,
.
.
.
,
ε
m
\varepsilon_1,\varepsilon_2,...,\varepsilon_m
ε1,ε2,...,εm 的过程如下
- example
注意要严格按照定义计算,比如求 ε 1 \varepsilon_1 ε1 时,其中算 ∥ β 1 ∥ = ( β , β ) \Vert\beta_1\Vert = \sqrt{(\beta,\beta)} ∥β1∥=(β,β) 要带入题目给出的內积公式,即 ( β , β ) = ∫ − 1 1 1 d x = 2 \sqrt{(\beta,\beta)} = \sqrt{\int_{-1}^11dx}=\sqrt{2} (β,β)=∫−111dx=2